Khalifa University of Science and Technology’s Digital Future Institute has launched RF-GPT, described as the first radio-frequency AI language model capable of interpreting wireless signals through natural language, addressing a fundamental gap in telecom AI where large language models have historically been limited to text and structured network data.
Machine Learning & Artificial Intelligence
RF-GPT converts radio signals into visual spectrograms that AI systems can then analyse and describe in plain language, enabling queries about wireless spectrum activity without specialist tooling. In benchmark testing, the model outperformed existing baseline models by up to 75.4% on radio frequency spectrogram tasks, and correctly counted the number of signals present in a spectrogram approximately 98% of the time — a task that general-purpose AI models almost never achieve. The model was trained on approximately 625,000 computer-generated radio signal examples and demonstrated strong performance across signal type identification, overlapping transmission detection, wireless standard recognition, Wi-Fi device estimation, and 5G signal extraction.
The project was led by Professor Merouane Debbah, Senior Director of the Digital Future Institute, with collaborators from Khalifa University, Université de Lorraine, and Zhejiang University. The university framed RF-GPT as a direct contribution to the UAE National Artificial Intelligence Strategy and positioned it as foundational infrastructure for AI-native 6G networks, where physical-layer spectrum data can be queried and acted upon in real time.
